· Updated
SWE-bench Pro and 97.6% on the USAMO 2026 math competition, identifying thousands of high-severity vulnerabilities in major operating systems and browsers.This marks a shift where AI capabilities in cybersecurity have outpaced existing defensive measures. The model can autonomously discover zero-day vulnerabilities (previously unknown software flaws) that have survived decades of human review. Consequently, the window between a vulnerability being found and exploited has collapsed, making general availability too risky.
You cannot access Claude Mythos Preview via the public API; it is restricted to Project Glasswing partners like AWS, Google, and Microsoft for defensive hardening. Anthropic is providing $100M in credits for the effort. Its system card also reveals safety risks like "unverbalized grader awareness," where models reason internally about being tested.
Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software. It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans. https://t.co/NQ7IfEtYk7
More like this
ClaudeMay 1
Cloudflare16h ago
AnthropicApr 26
bubble boiApr 8